The use of artificial intelligence as a tool to assist judges in the sentencing process has attracted growing attention in contemporary legal scholarship. Nevertheless, the practical implementation of this technology within existing ethical and political frameworks—particularly in non-ideal penal systems—has received comparatively limited examination. This study seeks to address this gap by drawing upon the recent work of Jesper Ryberg and exploring the feasibility of employing algorithmic decision-support systems in criminal justice.The analysis is based on two foundational assumptions. The first is the overpunishment hypothesis, which maintains that many criminal justice systems impose punishments that exceed what offenders morally deserve. The second is the status quo preservation hypothesis, according to which the institutional acceptance of algorithmic systems depends on their ability to operate without causing systematic disruption to the existing penal order. In response to the challenges posed by these assumptions, the study examines the Restricted Application Model as a four-stage operational framework for the use of algorithmic sentencing support.
The originality of this research lies in its critical evaluation of this model and its assessment of both the opportunities and limitations associated with the gradual integration of decision-support algorithms into criminal justice systems. The findings suggest that the ethical legitimacy and practical feasibility of algorithmic sentencing depend not on replacing judicial discretion, but on carefully constraining the role of artificial intelligence within a human-centered decision-making framework.